fix: DeepSeek niche/type not saving to DB

Two bugs:
1. _parse_classify_output stripped <think> block before searching for JSON.
   DeepSeek-R1 often puts the JSON array inside the think block (especially
   when it "decides" mid-reasoning), so stripping it first destroyed the data.
   Fix: search full output first, then inside <think>, then stripped — three
   fallback strategies with info logging at each step.

2. Phase 2 save used bare UPDATE WHERE domain=? which silently does nothing
   if the domain row doesn't exist yet in enriched_domains.
   Fix: replace with INSERT ... ON CONFLICT DO UPDATE (true upsert).

Also adds logger.info lines so container logs show raw DeepSeek output
and parse result count for easy debugging.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

app/db.py

@@ -436,10 +436,14 @@ async def save_prescreen_results(results: list[dict]):
             niche     = r.get("niche")
             site_type = r.get("type")   # DeepSeek returns "type" key
             if niche or site_type:
-                # Classification-only update (domain row must already exist)
+                # Upsert niche/type — works even if the row was never enriched
                 await db.execute(
-                    "UPDATE enriched_domains SET niche=?, site_type=? WHERE domain=?",
+                    """INSERT INTO enriched_domains (domain, niche, site_type)
-                    (niche, site_type, domain),
+                       VALUES (?, ?, ?)
+                       ON CONFLICT(domain) DO UPDATE SET
+                         niche=excluded.niche,
+                         site_type=excluded.site_type""",
+                    (domain, niche, site_type),
                 )
             else:
                 # Prescreen status upsert — create row if it doesn't exist yet

app/prescreener.py

@@ -175,16 +175,48 @@ def _build_classify_prompt(items: list[dict]) -> str:
 
 
 def _parse_classify_output(raw: str) -> list[dict]:
+    """Extract JSON array from DeepSeek output.
+    Strategy: search the full raw text first (handles cases where the JSON
+    sits outside or inside <think> blocks), then try with think stripped.
+    DeepSeek-R1 sometimes puts its answer inside the think block; stripping
+    it first would lose the data entirely.
+    """
     text = re.sub(r"```(?:json)?", "", raw).strip().rstrip("`").strip()
-    # Strip DeepSeek <think>…</think> reasoning block if present
+
-    text = re.sub(r"<think>[\s\S]*?</think>", "", text).strip()
+    def _try_parse(s: str):
-    m = re.search(r"\[[\s\S]+\]", text)
+        m = re.search(r"\[[\s\S]+\]", s)
-    if m:
+        if not m:
+            return None
         try:
-            return json.loads(m.group(0))
+            result = json.loads(m.group(0))
+            if isinstance(result, list) and result:
+                return result
         except json.JSONDecodeError:
             pass
-    logger.warning("DeepSeek classification parse failed: %s", raw[:300])
+        return None
+
+    # 1. Try the full output as-is (handles JSON after </think>)
+    parsed = _try_parse(text)
+    if parsed:
+        logger.info("DeepSeek: parsed %d items from full output", len(parsed))
+        return parsed
+
+    # 2. Try ONLY the content inside <think> (handles JSON inside the block)
+    think_m = re.search(r"<think>([\s\S]*?)</think>", text)
+    if think_m:
+        parsed = _try_parse(think_m.group(1))
+        if parsed:
+            logger.info("DeepSeek: parsed %d items from <think> block", len(parsed))
+            return parsed
+
+    # 3. Try with think block stripped (standard path)
+    stripped = re.sub(r"<think>[\s\S]*?</think>", "", text).strip()
+    parsed = _try_parse(stripped)
+    if parsed:
+        logger.info("DeepSeek: parsed %d items after stripping <think>", len(parsed))
+        return parsed
+
+    logger.warning("DeepSeek classification parse failed, raw snippet: %.400s", raw)
     return []
 
 
@@ -217,7 +249,10 @@ async def classify_with_deepseek(live_items: list[dict]) -> list[dict]:
         if isinstance(output, list):
             output = "".join(output)
 
-        return _parse_classify_output(output)
+        logger.info("DeepSeek raw output (first 500 chars): %.500s", output)
+        result = _parse_classify_output(output)
+        logger.info("DeepSeek classified %d / %d domains", len(result), len(live_items))
+        return result
 
     except Exception as e:
         logger.error("DeepSeek classification error: %s", e)